Statistical parameters for assessing environmental model performance related to sample size: Case study in ocean color remote sensing
share
Environmental model performances need to be assessed using some statistical parameters, such as mean absolute error (MAE) and root mean square error (RMSE). The advantages and disadvantages of these parameters are still in controversial. The purpose of this study is to introduce a statistical parameter, type A uncertainty (UA), into model performance evaluations. We particularly focus on the relations between sample sizes and three evaluation parameters, and tested a few ocean color remote sensing algorithms and datasets. The results indicate that RMSE, MAE and UA all vary with the sample size n but present different trends. Based on our tested results and theoretical analysis, we therefore conclude that UA is better than RMSE and MAE to express model uncertainty, because its downward trends indicate that the more samples we take, the less uncertainty we get. RMSE and MAE are good parameters for assessing model accuracy rather than uncertainty.
READ FULL TEXT
